Revert "halucination fixes"

halucination fixes

prompt

prompt change

prompt

prompt

prompt optimization

user friendly prompt change

fixes for formatted output

prompt update

update prompt

Modified Dockerfile to include build-essential and cmake.

docker and requirements.txt

pinned torch==2.3.1, torchvision==0.18.1, and torchaudio==2.3.1 in requirements.txt to resolve the version conflict.

removed openai-whisper from requirements.txt.

RUN pip install --no-cache-dir --no-build-isolation openai-whisper==20231117

fixed the build error related to openai-whisper. It failed because it relies on pkg_resources, which was removed in recent versions of setuptools. I've updated the Dockerfile to use a compatible version (setuptools<70.0.0).

preload disabled

entrypoint.sh !models_config.json in docker ignore

fix for the "file not found" error

docker files fix

Merge branch 'FT-DEV-20/01/2026'

rag

Revert "refactor(ai-service): optimize prompts, token counting, and benchmarking - Externalize system prompts to text files for better maintainability. - Integrate tiktoken for faster and more accurate token counting. - Refactor BenchmarkLogger to use asynchronous logging for zero latency impact. - Improve prompt echo removal logic with more robust markers. - Add specialized system instruction for medical document processing. - Update test expectations to reflect precise token counting."

Revert "update path for verify_cache.py"

Revert "disable preload models"

Revert "minimal docker"

Revert "use existing entrypoint.sh instead of inline generation""

Revert "entry point docker ignore"

entry point docker ignore

use existing entrypoint.sh instead of inline generation"

minimal docker

disable preload models

update path for verify_cache.py

refactor(ai-service): optimize prompts, token counting, and benchmarking - Externalize system prompts to text files for better maintainability. - Integrate tiktoken for faster and more accurate token counting. - Refactor BenchmarkLogger to use asynchronous logging for zero latency impact. - Improve prompt echo removal logic with more robust markers. - Add specialized system instruction for medical document processing. - Update test expectations to reflect precise token counting.

class 'generator' error.

traceback in benchmark api

error success benchmark

exceptions and error logs added

bench mark

update unified manager.py

resolve streaming generation errors and improve robustness

dynamic cache fix

dynamic cache patch

openvino fix

changes for publishing the latest including generate_generic api

Optimize performance and fix torchvision error

Local running issue fix

Update requirements.txt

Environment variables issue fix